忘记那根棍子：通过电路归因实现量化-永久反学习
Forgetting That Sticks: Quantization-Permanent Unlearning via Circuit Attribution
AI总结：
本文提出MANSU方法，通过电路归因隔离最小遗忘子图，结合电路受限的空域投影和幅度下限保证量化生存，解决量化压缩下的遗忘与保留问题，同时引入CAD指标区分结构擦除与行为抑制。
摘要：
标准反学习评估在全精度下测量行为抑制，尽管所有部署的语言模型首先进行量化。近期研究表明4位后训练量化可以逆转机器反学习；我们证明这不是调谐的副作用，而是系统性的双重失败：基于梯度的方法在压缩下失去有意义的遗忘，而能承受量化的方 法几乎不改变模型。这两种失败都源于同一个根本原因：所有基线中，每个参数的更新量低于NF4量化桶宽度47-828倍；更新扩散到数十亿参数无法清除量化桶边界，我们将其正式化为稀疏性-持久性权衡。我们提出MANSU（机理对齐的空域反学习），通过结合因果电路归因隔离最小遗忘子图，电路受限的空域投影与对角Fisher保留界，以及每参数幅度下限保证量化生存。我们还引入电路归因发散（CAD），一个机理验证度量，区分结构擦除与行为抑制，这一区分现有度量无法实现。在多个模型家族和危险基准上，MANSU是第一个同时满足所有四个属性的方法，每个属性都有边际优势（有意义的遗忘、保留保护、非正PTQ间隙和结构擦除），而基于梯度的基线在压缩下恢复高达+0.05准确率。
英文摘要：
Standard unlearning evaluations measure behavioral suppression in full precision, immediately after training, despite every deployed language model being quantized first. Recent work has shown that 4-bit post-training quantization can reverse machine unlearning; we show this is not a tuning artefact but a systematic dual failure: gradient-based methods that achieve meaningful forgetting lose it under compression, while methods that survive quantization barely change the model. Both failures trace to the same root cause: across all baselines, per-parameter updates lie 47-828x below the NF4 quantization bin width; updates diffused across billions of parameters cannot clear quantization bin boundaries, a consequence we formalize as a sparsity-permanence tradeoff. We present MANSU (Mechanistic-Aligned Null-Space Unlearning), which resolves both modes by combining causal circuit attribution to isolate the minimal forget-set subgraph, circuit-restricted null-space projection with a diagonal-Fisher retain bound, and a per-parameter magnitude floor guaranteeing quantization survival by construction. We additionally introduce Circuit Attribution Divergence (CAD), a mechanistic verification metric distinguishing structural erasure from behavioral suppression, a distinction existing metrics cannot make. Across multiple model families and hazard benchmarks, MANSU is the first method to jointly satisfy all four properties with margin on each (meaningful forgetting, retain preservation, non-positive PTQ gap, and structural erasure), while gradient-based baselines recover up to +0.05 accuracy under compression.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
